Similarity engine for content-based retrieval of images
A system and method for content-based search and retrieval of visual objects. A base visual information retrieval (VIR) engine utilizes a set of universal primitives to operate on the visual objects. An extensible VIR engine allows custom, modular primitives to be defined and registered. A custom primitive addresses domain specific problems and can utilize any image understanding technique. Object attributes can be extracted over the entire image or over only a portion of the object. A schema is defined as a specific collection of primitives. A specific schema implies a specific set of visual features to be processed and a corresponding feature vector to be used for content-based similarity scoring. A primitive registration interface registers custom primitives and facilitates storing of an analysis function and a comparison function to a schema table. A heterogeneous comparison allows objects analyzed by different schemas to be compared if at least one primitive is in common between the schemas. A threshold-based comparison is utilized to improve performance of the VIR engine. A distance between two feature vectors is computed in any of the comparison processes so as to generate a similarity score.
Latest Virage, Inc. Patents:
Claims
1. A search engine, comprising:
- a function container capable of storing primitive functions;
- a registration interface storing functions to the function container; and
- a primitive supplying primitive functions to the registration interface, wherein the primitive functions include an analysis function capable of extracting features from an object.
2. The engine defined in claim 1, wherein the primitive functions further comprise a comparison function to compare visual objects.
3. A search engine, comprising:
- a set of primitives each having one or more functions capable of extracting attributes from a visual object and capable of determining similarity between visual objects;
- a registration interface capable of registering the primitives; and
- a comparator function capable of applying functions of registered primitives so as to compare objects.
4. The engine defined in claim 3, further comprising an analysis function capable of deriving feature data from the object.
5. In a system comprising a search engine in communication with a database including a plurality of visual objects, a method of object comparison, comprising the steps of:
- creating a custom primitive having an analysis function and a comparison function;
- registering the custom primitive with the search engine via a registration interface;
- applying the analysis function to at least a portion of the database objects and a query object; and
- applying the comparison function to the analyzed query object and at least a portion of the analyzed database objects.
6. The method defined in claim 5, additionally comprising the step of storing the analysis function and the comparison function in a function container.
7. The method defined in claim 5, additionally comprising the step of defining a custom primitive for extracting an attribute from an object.
8. The method defined in claim 5, additionally including the step of defining a schema of primitives for a specific domain of objects.
9. The method defined in claim 5, additionally including the step of assigning a primitive identification (ID) tag to the custom primitive.
10. The method defined in claim 7, wherein each object comprises an image.
11. The method defined in claim 10, wherein the attribute is extracted over the entire image.
12. The method defined in claim 10, wherein the attribute is extracted over only a portion of the image.
13. A method of extending a search engine for searching a database of visual objects, the method comprising the steps of:
- providing a search engine primitive including a function to extract an attribute from a visual object;
- adding a new primitive to the search engine via a registration interface, wherein the new primitive comprises a comparison function for a selected attribute; and
- comparing the extracted attributes from two visual objects with the comparison function.
14. The method defined in claim 13, wherein the adding step comprises subclassing and function overloading of the primitive.
| 4672683 | June 9, 1987 | Matsueda |
| 5148522 | September 15, 1992 | Okazaki |
| 5515453 | May 7, 1996 | Hennesey et al. |
| 5579471 | November 26, 1996 | Barber et al. |
| 5729704 | March 17, 1998 | Stope et al. |
- A Cognitive Approach to Visual Interaction, T. Kato, et al., International Conference on Multimedia Information Systems '91, 1991, pp. 109-120. Using Depicitive Queries to Search Pictorial Databases, S. Charles, et al., Interact '90 Proceedings of the IFIP TC 13 In: Human-Computer Interaction, 1990, pp. 493-498. Ultimedia Manager: Professional Edition for OS/2 & DB2/2 brochure, IBM. IBM Unleashes QBIC Image-Content Search, The Seybold Report on Desktop Publishing, Sep. 12, 1994, pp. 34-35. Hands on Information: Visualizer Ultimedia Query for OS/2 brochure, IBM, 1994. Database Architecture for Content-Based Image Retrieval, T. Kato, SPIE vol. 1662 Image Storage and Retrieval Systems, 1992, pp. 112-123.
Type: Grant
Filed: Mar 28, 1997
Date of Patent: Apr 6, 1999
Assignee: Virage, Inc. (San Mateo, CA)
Inventors: Ramesh Jain (San Diego, CA), Bradley Horowitz (San Mateo, CA), Charles E. Fuller (San Mateo, CA), Amarnath Gupta (Redwood City, CA), Jeffrey R. Bach (San Mateo, CA), Chiao-fe Shu (San Mateo, CA)
Primary Examiner: Wayne Amsbury
Assistant Examiner: Thu-Thao Havan
Law Firm: Knobbe, Martens, Olson & Bear, LLP
Application Number: 8/829,791
International Classification: G06F 1730;
